Dev builds AI side panel Chrome extension in two weekends, struggles with distribution

·1 views

A developer built and shipped a Chrome extension in roughly 30 hours across two weekends, allowing users to highlight text on any webpage and receive AI-generated replies in a persistent side panel. The extension supports four AI models — OpenAI, Claude, DeepSeek, and Gemini — with the user's API key stored locally and no backend required. A key technical hurdle was a Content Security Policy issue that prevented direct API calls from the side panel, requiring a workaround using Chrome's background service worker and message-passing system. After Chrome Web Store approval and posts on Reddit, X, and Indie Hackers, the extension attracted only 15 users over five weeks. The developer concluded that while rapid building is achievable, user distribution remains the harder and more consequential challenge.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Developer builds zero-dependency browser-based JSON Schema Draft-07 validator

A developer has released a browser-only JSON Schema validator that requires no external libraries or Node.js setup. The tool implements a broad subset of the Draft-07 specification, covering keywords for type checking, string and number constraints, array and object validation, combiners like allOf, anyOf, and oneOf, conditionals, and local $ref references. Built as a single recursive function, it returns structured errors that include the JSON path, a human-readable message, and the specific failing keyword. The project was designed to make quick schema debugging and learning more accessible without spinning up a full development environment. The live tool is publicly available and has been validated against 173 tests.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Five Signals That Show a Business Workflow Is Ready for AI Automation

Many companies have inefficient workflows that employees quietly work around, often managed through spreadsheets, Slack threads, and tribal knowledge rather than formal systems. According to a framework developed from over a decade of AI and software development experience, a workflow becomes a strong candidate for AI automation when it exhibits multiple readiness signals. These signals include high repetition, repeated human judgment calls, manual data movement between tools, delays caused by waiting for approvals or context, and a measurable impact on cost, revenue, or customer experience. Workflows showing three or more of these signals — such as a spreadsheet-driven customer onboarding process spanning CRM, Slack, and billing tools — are described as prime targets for redesign. The core argument is that AI automation should begin as a workflow investigation rather than a technology project, since attaching AI to a poorly designed process only accelerates its failure.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Pinecone, Qdrant, Weaviate, Milvus: How to Pick the Right Vector DB in 2026

A hands-on analysis of four leading vector databases — Pinecone, Qdrant, Weaviate, and Milvus — highlights how each suits different production constraints rather than one being universally superior. Pinecone offers a fully managed, zero-infrastructure option best suited for datasets under 10 million vectors, while Qdrant stands out for filtering quality, native hybrid search, and lowest cost at scale. Weaviate brings built-in vectorization and mature hybrid search capabilities, and in April 2026 became the first vector database with a native MCP Server for direct LLM and agent integration. Milvus is the recommended choice for datasets exceeding 100 million vectors, leveraging GPU-accelerated indexing and a Kubernetes-based architecture, and recently replaced its messaging dependency with its own WAL system called Woodpecker. Benchmarks from Salt Technologies AI's 2026 report show Qdrant achieving the lowest latency at 4ms p50, while Pinecone's serverless cold queries can spike to 40–80ms — a gap that can significantly impact applications with bursty traffic patterns.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AI-Generated Godot Code Works Until It Doesn't — The Hidden Maintainability Problem

Developers using Claude Code to generate GDScript for Godot 4.x games can produce working code quickly, but a pattern dubbed 'Skeleton Implementation' is emerging as a long-term risk. The AI generates structurally correct code — classes, functions, scene logic — without transferring the underlying spatial and architectural reasoning to the developer. A detailed Qiita post by a Japanese indie developer, documenting AI-assisted title screen implementation using Claude Code and MCP, illustrates how this plays out in practice. While the generated code handles standard cases like scene transitions and viewport scaling, developers struggle to extend or debug it when edge cases arise — such as camera state persistence or dynamic resolution changes — because they never built the mental model behind the code. The concern is that AI-assisted codebases may pass all tests and appear functional, yet become unmaintainable even by the original developer within weeks.

0 comments Read more at DEV Community